Modularity in the decentralized control of large scale systems

Large scale systems have complex models which have high order and many inputs and outputs. Typically such systems are composed of many interconnected subsystems which are geographically separated. A control agent is associated with the inputs and outputs of each subsystem. This thesis is concerned with the manner in which modularity can be introduced to simplify the analysis and controller design of such systems. In particular, this work explores modular system construction, servocompensator modularity, modular model reduction, and the decomposition of system dynamics through the use of time scale techniques for analysis and controller design.;Conditions that allow a modular construction of a system are derived. Suppose a system exists with the property that decentralized controllers have been applied to solve the robust decentralized servomechanism problem (RDSP) so that tracking and regulation occur in each of the control agents. Suppose a new subsystem which is to be connected to the original subsystem also has a set of tracking and regulation requirements. Conditions are derived to ensure that such an interconnection can be accomplished without requiring adjustment of the existing controllers. Only the new controllers need be adjusted.;Necessary and sufficient conditions that ensure the existence of a solution to the RDSP, for the case when control agents have distinct tracking and regulation requirements, are derived. It is found that the decentralized servocompensator is a modular composition of components, each of which is tailored to the requirements of a different subsystem.;For systems composed of a large number of subsystems, some form of model reduction is required for both controller design and simulation. A direct approach requires that detailed models of each subsystem be interconnected to produce a high order composite model which is then simplified. The intermediate use of this high order composite model is undesirable for large systems. In this work, conditions which ensure that simplified subsystem models can be interconnected to yield results consistent with dominant time dynamics of the full order system are derived. This allows a modular approach to modelling to be used.;Time scale techniques are used to decompose the existence conditions to the RDSP. In contrast to earlier work, a formulation which is consistent with output feedback design methods is investigated. Implications of information transmission between control agents are also investigated. This decomposition is used to simplify the parameter optimization required in the design of decentralized controllers that solve the RDSP.